Consequences of Measurement Problems in Strategic Management Resea

8/14/2019 Consequences of Measurement Problems in Strategic Management Resea

1/9

Strategic Management JournalStrat. Mgmt. J., 26: 367375 (2005)

Published online 22 December 2004 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/smj.445

RESEARCH NOTES AND COMMENTARIES

CONSEQUENCES OF MEASUREMENT PROBLEMS IN

STRATEGIC MANAGEMENT RESEARCH: THE CASE

OF AMIHUD AND LEV

BRIAN K. BOYD,1* STEVE GOVE2 and MICHAEL A. HITT31 W. P. Carey School of Business, Arizona State University, Tempe, Arizona, U.S.A.2 Management/Marketing Department, University of Dayton, Dayton, Ohio, U.S.A.3 Mays Business School, Texas A&M University, College Station, Texas, U.S.A.

Strategic management research has been characterized as placing less emphasis on constructmeasurement than other management subfields. To illustrate the consequences of measurementerror, we revisit the debate on the causes of diversification. Our research suggests that thedivergentfindings between studies on this topic are largely the result of measurement error, andthat prior work has underestimated the true effect of size in the relationships between variables.Copyright 2004 John Wiley & Sons, Ltd.

Strategic management is generally acknowledged

to be one of the younger subdisciplines within

the broader management domain. Such emergent

areas are typically characterized by debate, and

challenges to existing paradigms (Kuhn, 1996).

While the latter are often couched as theoretical

discussions, empirical work plays a critical role in

confirming, or challenging, a particular perspec-

tive. Contributing to the advancement of the field,

there has been a small research stream that cri-

tiques empirical research in strategic management.

Regardless of the topic, these reviews have been

consistently critical of the rigor of strategic man-

agement research.

Keywords: measurement; research design; Type II error;agency theory; diversification; corporate governance*Correspondence to: Brian K. Boyd, W. P. Carey School ofBusiness, Arizona State University, Tempe, AZ 85287-4006,U.S.A. E-mail: [email protected]

Construct measurement is a key area of concern

for strategic management research, as the variables

of interest tend to be complex or unobservable

(Godfrey and Hill, 1995). Paradoxically, measure-

ment has been a low-priority topic for strategic

management scholars (Hitt, Boyd, and Li, 2004).

As a result, complex constructs have often been

represented with simple measures, and with limited

testing for reliability or validity (Boyd, Gove, and

Hitt, 2005). To illustrate the consequences of mea-

surement issues, we replicate a prominent debate

among strategy researchers regarding whether ornot diversification is a consequence of agency costs

(Amihud and Lev, 1981). Using data from 640

Fortune firms, we created multiple indicator mod-

els of both agency costs and diversification. Our

results provide strong evidence that the debate

between authors is largely an artifact of measure-

ment error.

Copyright 2004 John Wiley & Sons, Ltd. Received 13 January 2003Final revision received 26 July 2004


2/9

368 B. K. Boyd, S. Gove and M. A. Hitt

LITERATURE REVIEW ANDHYPOTHESES

A common explanation for diversification is the

continued search for growth. A mature firm might

consider expanding the scope of its offerings in

pursuit of new growth opportunities. An alternative

explanation is based in agency theory. Much as

investors strive to balance their personal portfolios

and thus their risk, agency theorists contend that

top managers expand the firms business portfolio

to mitigate their individual risk even if doing

so ultimately results in a reduction of shareholder

wealth.

Evidence suggests that the unique interests of

managers, including natural inclinations toward

risk aversion (Berle and Means, 1932; Jensen and

Meckling, 1976), help to explain many organiza-

tional phenomena including executive perquisites

(e.g., Boyd, 1994), governance innovations (e.g.,

Hoskisson and Hitt, 1994), and strategic initiatives

(e.g., Baysinger, Kosnik, and Turk, 1991; Sirower,

1997), among others.

The agency rationale has achieved the sta-

tus of conventional wisdom in the two decades

since Amihud and Levs (1981) seminal article.

Their study revealed that management-controlled

firms engaged in conglomerate mergers at a fargreater rate than owner-controlled organizations.

Because conglomerates are typically valued at a

discountmuch to the disadvantage of sharehold-

ers (Berger and Ofek, 1995; Denis, Denis, and

Sarin, 1997), Amihud and Lev (1981) concluded

that managerial self-interest is a primary motivator

behind diversification.

Relevance of Amihud and Lev to measurement

issues

Three factors guided our selection of Amihud

and Levs work to illustrate the consequences

of measurement error. First, while their results

have been largely accepted in the field, their work

was recently challenged. Second, there are issues

surrounding the measurement of both predictor and

dependent variables. Third, statistical power and

attenuation play a role in interpreting the results

to date. Next, we discuss each of these issues in

more detail.

Challenges to conventional wisdom

Debate and challenges to conventional wisdom are

central to a fields advancement (Kuhn, 1996).

Recently, Lane, Cannella, and Lubatkin (1998)

reanalyzed the Amihud and Lev data, and con-

cluded that owner monitoring had little effecton corporate diversification strategies. The debate

between these researchers was highlighted in a

recent issue of SMJ. Denis and colleagues sum-

marized the matter, noting that:

Though both sets of authors conduct similar empir-ical tests on virtually identical data, they arrive atcompletely different conclusions. Lane et al. (1999:1077) conclude that . . . there is little theoreticalor empirical basis for believing that monitoringby a firms principals influences its diversifica-tion strategy and investment decisions. In con-trast, Amihud and Lev (1999: 1064) conclude thatThe evidence shows that there exists a relation-ship between corporate diversification and corpo-rate ownership structure. (Denis, Denis, and Sarin,1999: 1071)

Measurement issues

Denis and colleagues (1999) argued that resolu-

tion of this debate hinges, in part, on a careful

evaluation of the empirical evidence. Their own

review suggested that the methodologies of both

studies were flawed, with an important shortfall

noted in the studies measurement approaches. For

example, each used broad ownership categories

constituting coarse-grained indicators of agency

conditions (e.g., McEachern, 1975; Palmer, 1973).

When improved constructs were substituted in the

analysesnamely, ratio-level indicators of equity

ownership, as well as refined measures of diversi-

ficationmore substantial results were generated

(Denis et al., 1997, 1999).

We believe that the confusion surrounding the

agencydiversification link is largely an artifact

of the methodologies used in studies, specifically

the measurement approaches. Empirical analysis

confirms that measurement error is more preva-lent for abstract vs. concrete concepts (Cote and

Buckley, 1987). Since the publication of Amihud

and Levs (1981) work, the fields understand-

ing of the key variables has advanced consid-

erably so, too, has our ability to measure the

specific variables of interest. In the context of

control alone, it is now well recognized that the

construct has several nuances (Fama and Jensen,

1983), leading researchers to recommend use of

multiple measures when studying control issues

(Eisenhardt, 1989). Recognizing the complexity of

measuring board oversight, one study developed a

Copyright 2004 John Wiley & Sons, Ltd. Strat. Mgmt. J., 26: 367375 (2005)


3/9

Research Notes and Commentaries 369

multi-indicator factor model to tap control (Boyd,

1994).

There are similar opportunities to refine the mea-surement of firm diversification. While there are

multiple measurement schemes available includ-

ing Rumelts categories and SIC countsthe

entropy measure (Palepu, 1985) has been reported

to have superior reliability and validity (Chatter-

jee and Blocher, 1992; Hoskisson et al., 1993).

The entropy measure is particularly germane to

our analysis, as it can be decomposed into unique

elementsindicators of both related and unrelated

diversification (Acar and Sankaran, 1999; Palepu,

1985).

Power

Of the core studies in this research stream, only

Lane and colleagues have explicitly addressed sta-

tistical power. They argued (Lane et al., 1998:

563) that their sample size of 309 had ample

power, as Cohen (1988: 13) observed that eco-

nomic research usually reports large effect sizes.

Additionally, they also suggested that their sample

had ample power to detect moderate effect sizes as

well. However, Cohen (1987) stated that the expec-

tation of large effect sizes may hold only when

using potent variables, and/or in the presence

of strong experimental controls. Separately, Cohen

(1987) also suggested that in noisy research a

moderate theoretical effect size may really end up

to be a small observed effect. Thus, differences

in expected effect sizes can dramatically change

the required sample size. Cohen (1992: 158) pro-

vided an example of a regression model with three

predictors, a significance level of p = 0.05, and

an 80 percent likelihood of identifying the rela-

tionship. The minimum sample size is 34 for a

large effect, 76 for a moderate effect, and 547

for a small effect. Lane et al. (1998) sampled 309firms, and Denis et al. (1997) sampled 933 firms.

Therefore, if there is a moderate theoretical effect

size between agency factors and diversification,

and measurement error exists, only Denis et al.

likely had sufficient power to capture an attenuated

effect.

The purpose of our study is to refine the

debate surrounding the control diversification

relationship. We build on the methodological

refinements recommended by Denis et al. (1997,

1999) and other scholars (e.g., Boyd, 1994;

Eisenhardt, 1989) to test a series of models

that use progressively more fine-grained measures

of both variables corporate control and extent

of diversification. Based on the prior theoreticalarguments offered in the previous studies of

these phenomena, we offer the following formal

hypotheses for testing:

Hypothesis 1: Board control is negatively related

to the level of diversification.

Hypothesis 2: The relationship between board

control and diversification is stronger when both

variables are measured with multiple indicators.

METHODS

Sample

Data were collected from a random sample of 640Fortune firms as part of a larger research project.

The sample included over 50 2-digit SICs, and

nearly 200 4-digit SICs. Company names were

selected randomly, and proxy statements were used

to collect governance data. Our design is cross-

sectional, with all data from the year 1987.

Analysis

In order to examine the effects of measurement

error and attenuation, we tested our hypotheses in

a structural model, using LISREL VII. Consistent

with the approach taken by Denis et al. (1997), we

used the extent of diversification as the dependent

variable, vs. merger activity. The model is shown

in Figure 1.

Measurement

Board control was measured using Boyds (1994)

multi-indicator factor model.1

The indicators forthis measure are CEO duality, ratio of insiders

to total board members, director stock ownership,

representation on the board by ownership groups,

1 Boyds model is not an exhaustive set of agency indicators.Thus, we conducted additional analyses to evaluate the robust-ness of our results. We developed new models that introduced asixth indicator, CEO tenure, as an additional measure of boardoversight. While tenure loaded on the board control factor model,its magnitude and level of statistical significance, while accept-able, were substantially less than the other extant indicators.Therefore, inclusion of a sixth indicator yielded only minorchanges in path coefficients, and tests of Hypotheses 1 and 2were unaffected.



4/9

370 B. K. Boyd, S. Gove and M. A. Hitt

Insiders

Owner Reps

Director Pay

CEO Duality

Assets

Sales

Equity

Stock Ownership

Firm Size

Diversification

Related

Unrelated

Board Control

-.43[5.9]

.38[5.3]

1.0

.99[10.2]

-.57[7.6]

.86[28.7]

.86[28.3]

1.0

-.16[2.2]

.11[2.3]

1.0

.63

[2.3]

Figure 1. Results of structural model. Note: Certain terms (e.g., theta and phi matrices) are omitted for ease ofrepresentation. t-values of parameters are noted in brackets; significance levels as follows: t= 2.0, p < 0.05; t= 2.7,

p < 0.01; t= 3.5, p < 0.001

and director pay. Proxy statements were used to

code these variables. CEO duality and director pay

loaded negatively on this construct, while the other

indicators loaded positively. Total diversification

(Palepu, 1985) was separated into its componentsdu (unrelated) and dr (related), using data from the

Compustat Business Segment database and com-

pany 10-K filings. Finally, we included firm size

as a control variable, because it has been pre-

viously linked to levels of diversification (Denis

et al., 1997). We measured size with three indica-

tors: net sales, total assets, and total stockholder

equity, also from Compustat. Log transformationswere used to normalize all size indicators.

RESULTS

Descriptive statistics for all variables are reported

in Table 1.

Tests of dimensionality

Prior to testing the hypotheses, we conducted a

series of analyses to confirm the factor loadings

and dimensionality of our predictor and control

variables. The first model represented a confirma-

tory factor analysis for the board control construct.

The results of this analysis are consistent with

Boyds (1994) results. All factor loadings were

in the expected direction, and statistically signifi-

cant at the p < 0.001 level. Overall fit measures

reported that a unidimensional model provided the

best fit to the data.

Second, we examined whether or not it is appro-

priate to treat dr and du as indicators of a com-

mon dimension. The full model (Figure 1) pro-

vides strong support for this assumption: dr wasused as the referent indicator, and the loading

for du was 0.63 (p < 0.01).2 However, an alter-

native argument could be made that the related

and unrelated diversification strategies are different

phenomena and, as such, likely have differing rela-

tionships with agency variables. For instance, man-

agers might consider related and unrelated portfo-

lios to have different types and levels of risk. If

2 Because there are only two indicators for this dimension, it isnot feasible to conduct a separate confirmatory factor analysisfor diversification.



5/9

Research Notes and Commentaries 371

Table 1. Descriptive statistics

du dr Sales Assets Equity Duality Dir.pay

Dir.equity

Ownerreps

Insiders

1. du 1.002. dr 0.12 1.003. Sales 0.12 0.16 1.004. Assets 0.01 0.06 0.69 1.005. Equity 0.07 0.17 0.80 0.80 1.006. Duality 0.10 0.07 0.06 0.02 0.06 1.007. Director pay 0.16 0.17 0.45 0.38 0.44 0.06 1.008. Director equity 0.09 0.09 0.20 0.24 0.25 0.26 0.21 1.009. Owner reps 0.07 0.05 0.15 0.20 0.23 0.19 0.23 0.52 1.00

10. Insiders 0.04 0.03 0.05 0.20 0.16 0.11 0.15 0.12 0.22 1.00X 0.29 0.15 7.47 7.63 6.48 0.79 21847 4.47 0.98 0.28 0.41 0.28 1.09 1.44 1.23 0.42 9163 11.52 1.60 0.14

Correlations greater than 0.08 significant at p < 0.05; values greater than 0.10 at p < 0.01.

true, dr and du would have unique associations

with ownership or monitoring variables. We tested

this competing perspective in a supplementary

model that treated dr and du as independent

constructs, and having separate paths from con-

trol and firm size i.e., a seemingly unrelated

regression. Using an incremental chi-square test,

this alternative model had a significantly worse

fit than the Figure 1 model. Our results provide

strong support for a multi-indicator approach tomeasuring diversification (as opposed to separate

measures of diversification types). Finally, fac-

tor loadings for the three size indicators were

highly statistically significant and in the expected

direction.

Model summary statistics

Coefficients were statistically significant and in the

expected direction for all structural and measure-

ment paths in Figure 1. Overall model measures

reported a very good fit: goodness of fit (GFI)was 0.94; the root mean square residual was 0.08;

other measures reported comparable fit. The coef-

ficient of determination, or R2, was 0.248 for the

dependent variables. In comparison, we explain

50 percent more variation of this variable than

Denis and colleagues (1997) analyses do, despite

using five fewer control variables. There was a sta-

tistically significant, negative covariation between

control and firm size (phi = 0.28, p < 0.001);

in other words, governance oversight tended to be

weaker in larger firms. Firm size has a positive

effect (0.11, p < 0.01) on diversification as well.

Hypothesis tests

Hypothesis 1 was supported with a statistically

significant, negative relationship ( = 0.16, p

Documents

Consequences of Measurement Problems in Strategic Management Resea